What are the best Topaz Video AI settings for upscaling AI-generated footage?

For AI-generated footage, start with Artemis HQ at a 2x pass, switch to Proteus in Manual (Fine Tune) when you hit AI-specific artifacts like plasticky skin or texture flicker, and use Starlight Precise for already-high-res Veo or Seedance 2.0 output. Keep sharpening low — AI footage is over-sharp by default, not soft.

Match the Topaz enhancement model to your generation source before touching any slider. Artemis HQ is the community-recommended universal starting point for low-to-medium quality input and works as the safe default for most AI clips. Proteus Fine Tune is the pick when you need granular manual control over synthetic artifact profiles — Runway or Kling clips with texture flicker or hallucinated detail bleed respond better to hand-set parameters than to any preset. Starlight Precise suits high-resolution Veo and Seedance 2.0 output, where the job is adding pixels rather than repairing them, and Iris works as a second enhancement pass when you need to recover fine facial detail. Topaz's own online upscaler now ships named presets for Kling 3.0, Seedance 2.0, and Veo — start from the preset that matches your generation model, preview, then build custom settings only if it falls short.

Auto vs Manual — the decision rule for AI clips. Auto mode reads the input's characteristics and sets parameters itself; it performs well when your source is clean and consistently generated. Switch to Manual (Proteus, using Estimate as a starting guide) when the footage shows problems Auto misreads: temporal flickering between frames, hallucinated texture bleed, or over-smoothed skin. These are synthetic-footage failure modes, not the compression and noise problems Auto was tuned around, so manual slider control wins on problem clips.

Keep sharpening and detail recovery low. Manual control matters most on these two sliders, because AI footage fails in the opposite direction of archival footage — it arrives over-sharp, not soft. As invideo's creative team puts it: "When you are generating a lot with seed dance, there tends to be this ultra-sharpness, there's this very plasticky feeling on the skin." Cranking sharpen or detail sliders amplifies exactly that artifact, so run noise reduction and recovery gently and avoid baking in extra sharpening or contrast — one documented production ran the upscale (Topaz Astra, which also runs directly on invideo) as the first post step, before blur, grain, and grade pulled the image back toward live action.

Resolution target. Set the output to your delivery format rather than the maximum available — 1080p to 4K covers most work, and Topaz supports targets up to 16K. A single 2x pass usually preserves more natural texture than an aggressive jump; preview a short section at your target before committing the full render.

These are solid starting points rather than fixed rules — the right settings depend on which model generated your footage and which artifacts it shows, so always preview before a full render.

Watch some of these to see what works for you:

When you are generating a lot with seed dance, there tends to be this ultra-sharpness, there's this very plasticky feeling on the skin.

— invideo's creative team